Merge bitcoin-core/secp256k1#1172: benchmarks: fix bench_scalar_split

eb6beba scalar: restrict split_lambda args, improve doc and VERIFY_CHECKs (Jonas Nick)
7f49aa7 ci: add test job with -DVERIFY (Jonas Nick)
620ba3d benchmarks: fix bench_scalar_split (Jonas Nick)

Pull request description:

  scalar_split_lambda requires that the input pointer is different to both output
  pointers. Without this fix, the internal benchmarks crash when compiled with
  -DVERIFY.

  This was introduced in commit bitcoin-core/secp256k1@362bb25 (which
  requires configuring with --enable-endomorphism to exhibit the crash).

  I tested that the new CI job would have caught this bug.

ACKs for top commit:
  sipa:
    utACK eb6beba
  real-or-random:
    utACK eb6beba

Tree-SHA512: c810545aefb01561ddb77b53618fa7acbb156ec13ab809c00523d4758492cafab1dfa01b6ebfb6195a3803bb49b16e63e8b0efcd1abb76ecefdb0476c3e483a3

.cirrus.yml

@@ -81,6 +81,7 @@ task:
     - env: {WIDEMUL: int128,                 ECDH: yes, SCHNORRSIG: yes}
     - env: {WIDEMUL: int128,  ASM: x86_64}
     - env: {                  RECOVERY: yes,            SCHNORRSIG: yes}
+    - env: {CTIMETESTS: no,    RECOVERY: yes, ECDH: yes, SCHNORRSIG: yes, CPPFLAGS: -DVERIFY}
     - env: {BUILD: distcheck, WITH_VALGRIND: no, CTIMETESTS: no, BENCH: no}
     - env: {CPPFLAGS: -DDETERMINISTIC}
     - env: {CFLAGS: -O0, CTIMETESTS: no}

src/bench_internal.c

@@ -110,10 +110,11 @@ static void bench_scalar_mul(void* arg, int iters) {
 static void bench_scalar_split(void* arg, int iters) {
     int i, j = 0;
     bench_inv *data = (bench_inv*)arg;
+    secp256k1_scalar tmp;
 
     for (i = 0; i < iters; i++) {
-        secp256k1_scalar_split_lambda(&data->scalar[0], &data->scalar[1], &data->scalar[0]);
-        j += secp256k1_scalar_add(&data->scalar[0], &data->scalar[0], &data->scalar[1]);
+        secp256k1_scalar_split_lambda(&tmp, &data->scalar[1], &data->scalar[0]);
+        j += secp256k1_scalar_add(&data->scalar[0], &tmp, &data->scalar[1]);
     }
     CHECK(j <= iters);
 }

src/scalar.h

@@ -88,9 +88,10 @@ static int secp256k1_scalar_eq(const secp256k1_scalar *a, const secp256k1_scalar
 
 /** Find r1 and r2 such that r1+r2*2^128 = k. */
 static void secp256k1_scalar_split_128(secp256k1_scalar *r1, secp256k1_scalar *r2, const secp256k1_scalar *k);
-/** Find r1 and r2 such that r1+r2*lambda = k,
- * where r1 and r2 or their negations are maximum 128 bits long (see secp256k1_ge_mul_lambda). */
-static void secp256k1_scalar_split_lambda(secp256k1_scalar *r1, secp256k1_scalar *r2, const secp256k1_scalar *k);
+/** Find r1 and r2 such that r1+r2*lambda = k, where r1 and r2 or their
+ *  negations are maximum 128 bits long (see secp256k1_ge_mul_lambda). It is
+ *  required that r1, r2, and k all point to different objects. */
+static void secp256k1_scalar_split_lambda(secp256k1_scalar * SECP256K1_RESTRICT r1, secp256k1_scalar * SECP256K1_RESTRICT r2, const secp256k1_scalar * SECP256K1_RESTRICT k);
 
 /** Multiply a and b (without taking the modulus!), divide by 2**shift, and round to the nearest integer. Shift must be at least 256. */
 static void secp256k1_scalar_mul_shift_var(secp256k1_scalar *r, const secp256k1_scalar *a, const secp256k1_scalar *b, unsigned int shift);

src/scalar_impl.h

@@ -52,7 +52,10 @@ static int secp256k1_scalar_set_b32_seckey(secp256k1_scalar *r, const unsigned c
  * nontrivial to get full test coverage for the exhaustive tests. We therefore
  * (arbitrarily) set r2 = k + 5 (mod n) and r1 = k - r2 * lambda (mod n).
  */
-static void secp256k1_scalar_split_lambda(secp256k1_scalar *r1, secp256k1_scalar *r2, const secp256k1_scalar *k) {
+static void secp256k1_scalar_split_lambda(secp256k1_scalar * SECP256K1_RESTRICT r1, secp256k1_scalar * SECP256K1_RESTRICT r2, const secp256k1_scalar * SECP256K1_RESTRICT k) {
+    VERIFY_CHECK(r1 != k);
+    VERIFY_CHECK(r2 != k);
+    VERIFY_CHECK(r1 != r2);
     *r2 = (*k + 5) % EXHAUSTIVE_TEST_ORDER;
     *r1 = (*k + (EXHAUSTIVE_TEST_ORDER - *r2) * EXHAUSTIVE_TEST_LAMBDA) % EXHAUSTIVE_TEST_ORDER;
 }
@@ -119,7 +122,7 @@ static void secp256k1_scalar_split_lambda_verify(const secp256k1_scalar *r1, con
  *
  * See proof below.
  */
-static void secp256k1_scalar_split_lambda(secp256k1_scalar *r1, secp256k1_scalar *r2, const secp256k1_scalar *k) {
+static void secp256k1_scalar_split_lambda(secp256k1_scalar * SECP256K1_RESTRICT r1, secp256k1_scalar * SECP256K1_RESTRICT r2, const secp256k1_scalar * SECP256K1_RESTRICT k) {
     secp256k1_scalar c1, c2;
     static const secp256k1_scalar minus_b1 = SECP256K1_SCALAR_CONST(
         0x00000000UL, 0x00000000UL, 0x00000000UL, 0x00000000UL,
@@ -139,6 +142,7 @@ static void secp256k1_scalar_split_lambda(secp256k1_scalar *r1, secp256k1_scalar
     );
     VERIFY_CHECK(r1 != k);
     VERIFY_CHECK(r2 != k);
+    VERIFY_CHECK(r1 != r2);
     /* these _var calls are constant time since the shift amount is constant */
     secp256k1_scalar_mul_shift_var(&c1, k, &g1, 384);
     secp256k1_scalar_mul_shift_var(&c2, k, &g2, 384);